docs

docs/build/html/_sources/install.rst

@@ -163,6 +163,8 @@ should point to the path to the built metal library.
      - ON
    * - MLX_BUILD_GGUF
      - ON
+   * - MLX_METAL_JIT
+     - OFF
 
 .. note::
 
@@ -196,9 +198,18 @@ GGUF, you can do:
   cmake ..
     -DCMAKE_BUILD_TYPE=MinSizeRel \
     -DBUILD_SHARED_LIBS=ON \
-    -DMLX_BUILD_CPU=ON \
+    -DMLX_BUILD_CPU=OFF \
     -DMLX_BUILD_SAFETENSORS=OFF \
-    -DMLX_BUILD_GGUF=OFF
+    -DMLX_BUILD_GGUF=OFF \
+    -DMLX_METAL_JIT=ON
+
+THE `MLX_METAL_JIT` flag minimizes the size of the MLX Metal library which
+contains pre-built GPU kernels. This substantially reduces the size of the
+Metal library by run-time compiling kernels the first time they are used in MLX
+on a given machine. Note run-time compilation incurs a cold-start cost which can
+be anwywhere from a few hundred millisecond to a few seconds depending on the
+application. Once a kernel is compiled, it will be cached by the system. The
+Metal kernel cache persists accross reboots.
 
 Troubleshooting
 ^^^^^^^^^^^^^^^

docs/build/html/_sources/python/_autosummary/mlx.core.gather_mm.rst

@@ -0,0 +1,6 @@
+mlx.core.gather\_mm
+===================
+
+.. currentmodule:: mlx.core
+
+.. autofunction:: gather_mm

docs/build/html/_sources/python/_autosummary/mlx.core.gather_qmm.rst

@@ -0,0 +1,6 @@
+mlx.core.gather\_qmm
+====================
+
+.. currentmodule:: mlx.core
+
+.. autofunction:: gather_qmm

docs/build/html/_sources/python/_autosummary/mlx.core.trace.rst

@@ -0,0 +1,6 @@
+mlx.core.trace
+==============
+
+.. currentmodule:: mlx.core
+
+.. autofunction:: trace

docs/build/html/_sources/python/ops.rst

@@ -35,7 +35,6 @@ Operations
    bitwise_or
    bitwise_xor
    block_masked_mm
-   block_sparse_mm
    broadcast_to
    ceil
    clip
@@ -69,6 +68,8 @@ Operations
    floor
    floor_divide
    full
+   gather_mm
+   gather_qmm
    greater
    greater_equal
    identity
@@ -149,6 +150,7 @@ Operations
    tensordot
    tile
    topk
+   trace
    transpose
    tri
    tril